sentiment analysis

25
Sentiment Analysis Applied Advertising & Public Relations Research JOMC 279

Upload: ull

Post on 23-Feb-2016

54 views

Category:

Documents


1 download

DESCRIPTION

Sentiment Analysis. Applied Advertising & Public Relations Research JOMC 279. "Listening is the study of naturally occurring conversations, behaviors, and signals—information that may or may not be guided—that brings the voice of people's lives in to a brand.". Why Do Brands Listen?. - PowerPoint PPT Presentation

TRANSCRIPT

Page 1: Sentiment Analysis

Sentiment Analysis

Applied Advertising & Public Relations Research

JOMC 279

Page 2: Sentiment Analysis
Page 3: Sentiment Analysis
Page 4: Sentiment Analysis
Page 5: Sentiment Analysis
Page 6: Sentiment Analysis
Page 7: Sentiment Analysis

"Listening is the study of naturally occurring

conversations, behaviors, and signals—information that may or may not be guided—that brings the

voice of people's lives in to a brand."

Page 8: Sentiment Analysis

Why Do Brands Listen?

• Insights (wants, unmet needs, challenges)• Voice of consumer• Redefine relationships• Understand shifts in perspectives• Understand context & reasons why

Page 9: Sentiment Analysis

Where Do Brands Listen?

• Offline– Comment cards– Trade-show notes– CRM / sales mgmt. systems

• Online– Brand backyard– Customer backyard

Page 10: Sentiment Analysis

Whom Do Brands Listen To?

• Customers• Prospects• Business partners• Friends, contacts, followers• Others

Page 11: Sentiment Analysis

How Do Brands Make Senseof What They Hear?

• Search & Monitoring• Text Analytics• Full-Service Listening Platforms• Private Communities

Page 12: Sentiment Analysis
Page 13: Sentiment Analysis

Measuring whatyour customers say about youwhen they're talking to each

other.

LISTENING

Page 14: Sentiment Analysis

Advantages (Online)

• Unobtrusiveness• Immediate / Real-time• Natural, rich, unfiltered WOM• BIG data

Page 15: Sentiment Analysis

Disadvantages (Online)

• Ethics• Representativeness / Accuracy• WOM Noise• BIG data

Page 16: Sentiment Analysis

Sentiment Analysis

• aka “opinion mining”• Measurement of emotion in texts– Polarity– Strength

• Human coding vs. NLP• Methodological standards / transparency

Page 17: Sentiment Analysis
Page 18: Sentiment Analysis
Page 19: Sentiment Analysis

Project 2 Results

• Data set: You were provided with 200 Tweets related to pizza. (2 sets)

• Code each Tweet as – Positive, Negative, Mixed, or Neutral.

• When coded as Positive, Negative, or Mixed, identify the portion of the Tweet that resulted in that decision.

• Evaluate the difficulty of the coding decision.

Page 20: Sentiment Analysis

Natural Language Processing

• SocialRadar vs. SentiStrength

• Observed agreement = .315 – Both data sets

• Why would computing kappa be inappropriate in this situation?

Page 21: Sentiment Analysis

OA kappaSentistrength 0.680 0.502Sentistrength 0.600 0.424Sentistrength 0.585 0.374Sentistrength 0.510 0.314Sentistrength 0.500 0.309Sentistrength 0.485 0.307Sentistrength 0.415 0.150

Social Radar 0.460 0.211Social Radar 0.450 0.151Social Radar 0.445 0.203Social Radar 0.400 0.207Social Radar 0.380 0.184Social Radar 0.365 0.188Social Radar 0.320 0.172

Page 22: Sentiment Analysis

“After coding these tweets, it is easy to see why computers might

not be the most effective way for a brand or company to decipher

customers’ tweets about a product or service.”

Page 23: Sentiment Analysis

“I have come to admire people who are professional coders.”

But are humans better?

Page 24: Sentiment Analysis

OA kappa0.660 0.4870.555 0.2950.525 0.310

0.810 0.7120.700 0.5560.670 0.5190.665 0.5130.665 0.5260.665 0.512

Page 25: Sentiment Analysis

Difficulty correlations

0.3970.3810.3580.3440.3380.2380.2290.1600.078